We often would like to compare a population parameter to our sample. We may know this parameter in advance, or we may want to compare it to another sample. Suppose we want to know for sure Subway is giving us a real footlong sandwich when we ask for one. We buy 30 12″ sandwiches and measure their lengths to be 11.5″ on average. Is this close enough to 12″ to accept Subway’s claim? Or we could have 30 samples from 2 different granites. Are these two granites geochemically equivalent or not?

In order to do this we follow a 5-step process

- Formulate a
**null hypothesis, H**(what we assume to be true), and our_{0}**alternative hypotheses, H**(which we only accept with sufficient evidence to reject the null)_{a} - Specify our level of significance. How important is it to avoid Type I errors. That is rejecting the null hypothesis, when it was in fact true. This could be very important when discussing the legal system and whether or not a person is innocent of a crime. It may be of less or equal importance in Subway sandwiches depending on your love of sandwiches.
- Calculate our test statistic.
- Define the region of rejection.
- Either accept our null hypothesis, or reject it in favour of our alternative hypothesis.

In our hypothesis testing, we can make two errors.

**Type I errors**occur when we reject our null hypothesis, but the null is true. For example, our justice system assumes a person is innocent. If we reject this, and put an innocent person in jail, we have made a type I error.**Type II errors**occurs when we accept the null, but our alternative was true. In this case, a guilty person gets away without punishment.

Although often type I errors are considered worse, this is not always true. Knowing the difference and the consequences for each error can impact your decisions. These errors are correlated, if we try to decrease the number of type II errors, the number of type I errors will increase.

This is the probability that the observed relationship between two sample sets is due to randomness, and that no difference exists between them. The ** p-value** helps us quantify this significance if it exists. It is a decreasing index of the reliability of the results. The smaller the p-value, the more reliable we believe that relation. Or in other words, a

*p*-value of 0.05 indicates that we have a 5% probability of rejecting our null hypothesis, when in fact it was correct (i.e. making a type I error). Or we have a 1/20 chance that our test would result in a relationship that is equal to or stronger than what we have observed.

We can also look at it as all hypothesis tests are in a sense “negative”. We can never 100% reject the null hypothesis, but we can disprove it to some level of significance. For example, **α = 0.05 (5% level of significance).**

A **power** of a test is the probability that our test rejects the null hypothesis, when the alternative hypothesis is true. We must specify the level of significance (α), before we do our test (or the probability of a type I error). If we want to minimize our type II errors (β), we express our null hypothesis with the intent to reject it. The power of a test is then 1-β.

Let’s do an example. Let’s assume we have 49 samples from a **normally distributed population**. We know σ = 14, but μ is unknown. We start with our hypothesis test, we suggest our population mean is 50, and test it against a one-sided alternative that the population mean is less than 50.

H_{0}: μ = 50H_{a}: μ < 50

We set our significance level at α = 0.05. We now pick our appropriate test statistic. We have a normal distribution, and we know our population standard deviation, the appropriate test statistic is:

We calculate which values of Z we reject our null hypothesis. We check our Z score from a Z table, and find that the Z value with an area of 0.05 to the left is -1.64. We therefore reject if Z ≤ -1.64. To facilitate our calculations of the power of the test, we express the rejection region as . We need to find what values of will we reject the null hypothesis. By expressing our Z and calculating as follows:

This means we have a 0.05 probability of committing a Type I error when our sample mean is 46.72 or less.

Now, if µ = 45, and our null hypothesis wrongly states µ = 50, what is the probability of a Type II error? That is, what’s the probability of not rejecting the null when it is in fact false.

We standardize as follows:

We look up in the Z table and find the area to the right of Z approximately 0.469. Therefore, our probability of a Type II error is β=0.469. And the probability of rejecting the null hypothesis when it is false, that is the **power of the test**, is 1 – β = 1 – 0.469 = 0.531!

If my explanations leave you with a few questions, which wouldn’t surprise me in the slightest, check out this jbstatistics video (or many of the other ones which are much better at showing graphics to explain the concepts and helped me understand them to begin with.

## Recent Comments